In this investigation, the welding quality of TC2 titanium alloy with 0.4 mm thickness was predicted using two regression models and an artificial neural network model. The welding current and the voltage between the upper and lower electrodes were obtained using the Rogowski coil and a line voltage sensor. And then the variations of the dynamic resistance curve and the effects of the welding current and welding time on the dynamic resistance signals were investigated. The principal component analysis (PCA) was employed to eliminate the redundant information in the dynamic resistance curve and characterize the shape information of the entire dynamic resistance. A linear regression model quantifying the relationship between the nugget diameter and the principal components was established. The results of the analysis of variance indicated that the performance of this regression equation was very good. Some statistical characteristics of the dynamic resistance signal were also extracted to investigate the relationship between the nugget diameter and dynamic resistance. The results indicated that the regression model established based on the PCA technique was much more robust than the model developed on the basis of the features manually extracted from the dynamic resistance signal. The neural network model was also used to predict the nugget diameter of the welding joints utilizing the extracted features. The performances of the three established prediction models were compared and their behavioral discrepancies were also investigated. The PCA technique not only can minimize the prior assumptions about the certain shape of the dynamic resistance curve and remove the subjective factors caused by the manual extraction method, but it also can assess and monitor the welding quality with a good level of reliability.

在这项研究中，使用两个回归模型和一个人工神经网络模型预测了厚度为0.4 mm的TC2钛合金的焊接质量。使用Rogowski线圈和线电压传感器获得上，下电极之间的焊接电流和电压。然后研究了动态电阻曲线的变化以及焊接电流和焊接时间对动态电阻信号的影响。主成分分析（PCA）用于消除动电阻曲线中的多余信息，并表征整个动电阻的形状信息。建立了线性回归模型，量化了熔核直径和主成分之间的关​​系。方差分析的结果表明，该回归方程的性能非常好。还提取了动电阻信号的一些统计特征，以研究熔核直径与动电阻之间的关系。结果表明，基于PCA技术建立的回归模型比基于从动态电阻信号手动提取的特征而开发的模型更加健壮。利用提取的特征，神经网络模型还用于预测焊接接头的熔核直径。比较了三个已建立的预测模型的性能，并研究了它们的行为差异。